[Bug 148807] [panic] "panic: sbdrop" and "panic: sbsndptr: sockbuf _ and mbuf _ clashing" (8.1-RELEASE/10.1-STABLE/11-CURRENT)

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=148807

Daniel Bilik <ddb at neosystem.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ddb at neosystem.org

--- Comment #32 from Daniel Bilik <ddb at neosystem.org> ---
(In reply to Robert Watson from comment #29)

> On the whole, my intuition is towards a device-driver bug based
> on past experience.

We've been also struggling this in past weeks, and I can confirm Robert's
intuition.

In our case, the bug affects two hosts running recent 10-STABLE, connected to
each other via igb(4) using a dedicated 100Mb switch. When trying to transfer
directory structure holding several gigabytes of data with rsync protocol,
either sender or receiver panics in less then a minute with:

Panic String: sbsndptr: sockbuf 0xfffff8000ccc76f8 and mbuf 0xfffff802a0145800
clashing

Interestingly, scp(1)ing data between the hosts doesn't seem to trigger this
panic such easily, but sometimes it does, mostly when copying larger (>1GB)
files.

We've fixed this just yesterday by limiting number of igb(4) txrx queues, ie.
adding this into loader.conf:

hw.igb.num_queues=1

Now the hosts run stable, periodically rsyncing data in both directions.

-- 
You are receiving this mail because:
You are the assignee for the bug.